Uncertainty in MRIO-based GHG footprints

15. Input-Output-Workshop Osnabrück

Simon Schulte*, Arthur Jakobs, Stefan Pauliuk

2024-03-01

Model uncertainty matters…

  • … for robust decision-making
  • … to guide GMRIO compilers to “uncertainty hot-spots” to allocate resources more efficiently

Uncertainty in EE-MRIO: Current state

  • database comparisons: Satellite accounts are largest source of discrepancy

https://robbieandrew.github.io/consumption/index.html

https://robbieandrew.github.io/consumption/index.html

Uncertainty in EE-MRIO: Current state

  • database comparisons: Satellite accounts are largest source of discrepancy
  • mostly points estimates, and, if at all, qualitative considerations of uncertainty
  • few studies provide quantitative estimation of parametric uncertainties [18]
  • two GMRIO databases that publish uncertainty estimates alongside each data entry (except for GHG extensions): Eora [9] & GLORIA [10]
  • all studies use Monte-Carlo (MC) simulations to propagate uncertainty from model input parameters to model outputs

Aim and scope of our study

  • estimate the parametric uncertainty of the GHG satellite accounts
  • estimate how the uncertainty propagates to GHG footprints
  • overcome two shortcomings from previous approaches:
    • #1: Uncertainty of the raw input data based on simplistic assumptions
    • #2: Correlations between variables obtained by disaggregating a common input data point are ignored
  • Scope:
    • Year 2015
    • GHGs CO\(_2\), CH\(_4\), N\(_2\)O
    • EXIOBASE country/sector resolution

Compiling GHG accounts

Statistical concept

Our workflow

Our workflow

Our workflow

Our workflow

Overcoming shortcomings of previous approaches

#1: Uncertainty of the raw input data based on simplistic assumptions

Two common approaches

Heuristics:

[2]

Statistical model: [1]

Our approach

UNFCCC National Inventory Reports (only in .pdf format 😭)

But now, as .csv on Zenodo: https://zenodo.org/records/10037714 😁

#2: A disregard of correlations between variables obtained by disaggregating a common input data point

Data disaggregation: Residence adjustment

Data disaggregation: Assign to MRIO sectors

Uncertainty propagation involving data disaggregation

G parallelogram1 Y₀ parallelogram2 Y₁ parallelogram1->parallelogram2 parallelogram3 Y₂ parallelogram1->parallelogram3 parallelogram4 Y₃ parallelogram1->parallelogram4

where \(\sum Y_i = Y_0\).

Uncertainty propagation involving data disaggregation

G parallelogram1 Y₀ parallelogram2 Y₁=α₁Y₀ parallelogram1->parallelogram2 parallelogram3 Y₂=α₂Y₀ parallelogram1->parallelogram3 parallelogram4 Y₃=α₃Y₀ parallelogram1->parallelogram4

where \(\sum \alpha_i = 1\).

Constraints & Information

G parallelogram1 Y₀ parallelogram2 Y₁=α₁Y₀ parallelogram1->parallelogram2 parallelogram3 Y₂=α₂Y₀ parallelogram1->parallelogram3 parallelogram4 Y₃=α₃Y₀ parallelogram1->parallelogram4
  1. Mean \(\mu\) and Standard Deviation \(\sigma\) of aggregate data (UNFCCC/EDGAR)
  2. Mean sector shares for disaggregate data \(\boldsymbol{\alpha}\) (proxy data: SUT/PEFA/…)
  3. sum-to-one constraint: \(\sum \alpha_i = 1\)
  4. no negative emissions

Sampling procedure

Step 1: Sample aggregate data from truncated normal distribution: \(y_0 \sim tN(\mu,\sigma,a = 0)\)

Step 2: Sample sector shares from Dirichlet distribution: \(x_1, x_2, ..., x_K \sim Dir(\boldsymbol\alpha,\gamma = \hat\gamma)\)

Step 3: Multiply both \(y_i = x_i y_0\ \forall\ i\in{1,..,K}\)

Sampling from a Dirichlet distribution

Sampling from a Dirichlet distribution

The \(\gamma\) parameter

\(\require{colorx} x_1, x_2, ..., x_K \sim Dir(\boldsymbol\alpha,\color{red} \gamma\color{black})\)

The \(\gamma\) parameter

\(\require{colorx} x_1, x_2, ..., x_K \sim Dir(\boldsymbol\alpha,\color{red}{\gamma=\hat\gamma} \color{black})\)

Maximum Entropy (MaxEnt) principle:
The least informative probability distribution consistent with a given set of constraints is the one which maximizes the entropy [11]

Finding \(\hat\gamma\) which maximises the entropy

\(\require{colorx} x_1, x_2, ..., x_K \sim Dir(\boldsymbol\alpha,\color{red}{\gamma=\hat\gamma} \color{black})\)

\[ \max_{\gamma>0}\ h(\gamma) = ln B(\gamma\boldsymbol\alpha) + (\gamma\alpha_0-K)\psi(\gamma\alpha_0) - \sum_{i=1}^K (\gamma\alpha_i-1)\psi(\gamma\alpha_i), \] where

  • \(\psi(x)\) is the Digamma function \(\psi(x) = \frac{d}{dn}ln(\Gamma(x)) = \frac{\Gamma'(x)}{\Gamma(x)}\),
  • \(\Gamma(x)\) is the Gamma function: \(\Gamma(x) = \int_0^\infty t^{x-1}e^{-t}dt\),
  • \(B(\gamma\boldsymbol\alpha)\) is the multivariate beta function: \(B(\gamma\boldsymbol\alpha) = \frac{\prod_{i=1}^K \Gamma(\gamma\alpha_i)}{\Gamma(\sum_{i=1}^K \gamma\alpha_i)}\)

Correlations

Why correlations matter

Why correlations matter

Results: Country level

Results: Sector level

Results: Correlations

Conclusion

  • Uncertainty hot-spots:
    • National level:
      • CO\(_2\): small economies subject to large residence adjustments
      • In general larger uncertainties for CH\(_4\) and esp. N\(_2\)O
    • Sector level: Overall: high uncertainties (median CV of ~1)
  • Ignoring correlations would overestimate CO\(_2\)-footprints and underestimate N\(_2\)O-footprints

Thank you!

References

1.
M. Lenzen, R. Wood, & T. Wiedmann, Uncertainty Analysis for Multi-Region InputOutput Models – a Case Study of the Uk’s Carbon Footprint. Economic Systems Research, 22 (2010) 43–63. https://doi.org/10.1080/09535311003661226.
2.
H. C. Wilting, Sensitivity and uncertainty analysis in mrio modelling; some empirical results with regard to the dutch carbon footprint. Economic Systems Research, 24 (2012) 141–171. https://doi.org/10.1080/09535314.2011.628302.
3.
J. Karstensen, G. P. Peters, & R. M. Andrew, Uncertainty in temperature response of current consumption-based emissions estimates. Earth System Dynamics, 6 (2015) 287–309. https://doi.org/10.5194/esd-6-287-2015.
4.
D. Moran, K. Kanemoto, M. Jiborn, R. Wood, J. Többen, & K. C. Seto, Carbon footprints of 13000 cities. Environmental Research Letters, 13 (2018) 064041. https://doi.org/10.1088/1748-9326/aac72a.
5.
P. Shrestha & C. Sun, Carbon emission flow and transfer through international trade of forest products. Forest Science, 65 (2019) 439–451. https://doi.org/10.1093/forsci/fxz003.
6.
H. Zhang, K. He, X. Wang, & E. G. Hertwich, Tracing the Uncertain Chinese Mercury Footprint within the Global Supply Chain Using a Stochastic, Nested InputOutput Model. Environmental Science & Technology, 53 (2019) 6814–6823. https://doi.org/10.1021/acs.est.8b06373.
7.
K. Kanemoto, Y. Shigetomi, N. T. Hoang, K. Okuoka, & D. Moran, Spatial variation in household consumption-based carbon emission inventories for 1200 Japanese cities. Environmental Research Letters, 15 (2020) 114053. https://doi.org/10.1088/1748-9326/abc045.
8.
K. Abbood, G. Egilmez, & F. Meszaros, Multi-region Input-Output-based Carbon and Energy Footprint Analysis of U.S. Manufacturing. Periodica Polytechnica Social and Management Sciences, 31 (2023) 91–99. https://doi.org/10.3311/PPso.19554.
9.
M. Lenzen, D. Moran, K. Kanemoto, & A. Geschke, Building eora: A global multi-region inputoutput database at high country and sector resolution. Economic Systems Research, 25 (2013) 20–49. https://doi.org/10.1080/09535314.2013.769938.
10.
M. Lenzen, A. Geschke, J. West, J. Fry, A. Malik, S. Giljum, L. Milà i Canals, P. Piñero, S. Lutter, T. Wiedmann, M. Li, M. Sevenster, J. Potočnik, I. Teixeira, M. Van Voore, K. Nansai, & H. Schandl, Implementing the material footprint to measure progress towards Sustainable Development Goals 8 and 12. Nature Sustainability, 5 (2022) 157–166. https://doi.org/10.1038/s41893-021-00811-6.
11.
E. T. Jaynes, Information Theory and Statistical Mechanics. Physical Review, 106 (1957) 620–630. https://doi.org/10.1103/PhysRev.106.620.